The present invention relates to a method and a device for determining an intermediary profile of a metal strip between an upstream and a downstream roll stand in a mill train.
In steel and aluminum manufacturing, optimization of the rolling process has a significant role both regarding the possible quality improvements and regarding cost reduction potential.
One important function of the process control of a mill train is determining the setting of the system for the next metal strip as accurately as possible before the strip enters the mill train. The process control is based on a series of models, which should describe the technological process taking place in the mill train as accurately as possible. This description is made difficult by the complexity of the process, which is affected by many non-linear influences and by changes in the process conditions over time.
The profile of the metal strip is an important quality criterion for its geometry after leaving the mill train. In the simplest case the profile is defined as the difference in thickness between the center and the edges of a metal strip. During the entire rolling process the profile of the strip entering a roll stand changes from one roll stand to the other. Since measuring the profile of the metal strip is complicated and costly, it is usually not measured until the end of the mill train and compared to the prediction of the model. Conventionally, in order to determine the profile (p1) of a metal strip downstream from the individual roll stands and thus ultimately the final profile, the following equation has been used repeatedly                               p          i                =                                            k              i                        ·                          p                              i                -                1                                      ·                                          h                i                                            h                                  i                  -                  1                                                              +                                    (                              1                -                                  k                  i                                            )                        ⁢                          Π              i                                                          (        1        )            
In this equation ki is calculated according to the article xe2x80x9cHigh Accuracy and Rapid-Response Hot Strip Mill,xe2x80x9d TECHNO Japan Vol. 20, No. 9, September 1987, pp. 54-59 using the formula                               k          i                =                                            1              π                        ⁢                          arctan              ⁡                              (                                                                            c                      i1                                        -                                          ln                      ⁢                                              xe2x80x83                                            ⁢                                              x                        i                                                                                                  c                    i2                                                  )                                              +                      1            2                                              (        2        )            
where                               x          i                =                                                                              D                  i                                ·                                  h                  i                  1.5                                                                    b              2                                .                                    (        3        )            
In addition,
The disadvantage here is that the equations for determining ki and xi apply only approximately. In addition, model parameters ci1 and ci2 are unknown and must be determined experimentally. This often results in insufficient determination of the final thickness profile.
This fact leads to the use of additional heuristic models. Experiments with adaptive neural networks have shown that these are capable of predicting the final profile in one step considerably more accurately than analytical models. Thus German Patent Application 196 42 918 describes a system for calculating the final profile of the metal strip, where selected parameters are determined using information processing based on neural networks.
An object of the present invention is to provide a method for determining intermediary profiles and a final profile of a metal strip in a mill train, according to which the intermediary profiles are determined more accurately than by conventional methods.
This object is achieved by providing a method and a device, where
at least one intermediary profile of a metal strip to be rolled is determined downstream from a roll stand and the final profile of this metal strip is determined after passing through all roll stands, using an analytical model for at least one roll stand;
this final profile is determined using a,second model, and
at least one parameter of the analytical model is modified so that the analytical model delivers a value for the final profile which coincides with the value obtained using the second model with a predefinable degree of accuracy.
The intermediary profiles are determined using an analytical model. In particular, an existing mathematical model of the process can be used whose parameters are optimized for each prediction. This expanded use of conventional models for a mill train considerably reduces the cost of the method according to the present invention and allows an existing mill train or roll stand to be retrofitted.
The second model is preferably a neural network in which non-linearities of the process can also be modeled. In particular, it is not used for optimizing the parameters of the analytical model until, after sufficient training, it delivers considerably better values for the final profile than the analytical model.